on firms’ product space evolution: the role of firm and

On firms’ product space evolution: the role of firmand local product relatednessAlessia Lo Turco*,y and Daniela Maggioni**

*Department of Economics and Social Sciences, Universita Politecnica delle Marche, Ancona, Italy**Department of Political and Social Science, University of Catania, Catania, Italy and Department ofEconomics and Social Sciences, Universita Politecnica delle Marche, Ancona, ItalyyCorresponding author: Alessia Lo Turco, Department of Economics and Social Sciences, UniversitaPolitecnica delle Marche, Ancona, Italy. [email protected]

AbstractWe explore the role of firm- and local product-specific capabilities in fostering theintroduction of new products in the Turkish manufacturing. Firms’ product spaceevolution is characterised by strong cognitive path dependence that, however, isrelaxed by firm heterogeneity in terms of size, efficiency and international exposure.The introduction of new products in laggard Eastern regions, which is importantly linkedto the evolution of their industrial output, is mainly affected by firm’s internal product-specific resources. On the contrary, product innovations in Western advanced regionshinge relatively more on the availability of local technological- related competencies.

Keywords: Product relatedness, firm heterogeneity, product innovationJEL classifications: D22, O53, O12

Date submitted: 18 March 2014 Date accepted: 31 May 2015

1. Introduction

A recent body of literature has highlighted the role of technological relatedness forstarting a new sector in a country (Hausmann and Hidalgo, 2009; Hidalgo, 2009), aregion (Boschma and Iammarino, 2009; Neffke et al., 2011; Boschma et al., 2012, 2013)or a firm (Breschi et al., 2003; Poncet and de Waldemar, 2012; Neffke and Henning,2013). In particular, empirical work in economic geography has shown that knowledgespillovers can spur innovation and growth across economic units located in space, aslong as the right—not too little, not too much (Nooteboom, 2000; Boschma, 2005)—extent of cognitive proximity exists among the relevant economic actors. Therefore, therole of geography in determining economic phenomena is complemented and enlivenedby the notion of technological relatedness. The latter also mediates the impact of firms’internal resources on the process of firm expansion. Some firm-level analyses in thefield, indeed, have stressed that firms tend to diversify in sectors that are technologicallyclose to their core activity (Breschi et al., 2003; Neffke and Henning, 2013).

The relevance of both within-firm and local cognitive proximities for innovationrecalls an important literature debate on the quantification of the absolute and relativecontribution of the existence of a fertile regional environment versus a firm’sendowment of suitable capabilities for product innovation (Pfirrmann, 1994;Sternberg and Arndt, 2001; Beugelsdijk, 2007; Wang and Lin, 2012).

� The Author (2015). Published by Oxford University Press. All rights reserved. For Permissions, please email: [email protected]

Journal of Economic Geography 16 (2016) pp. 975–1006 doi:10.1093/jeg/lbv024Advance Access Published on 2 July 2015

at Universita degli Studi di B

ergamo on January 2, 2017

http://joeg.oxfordjournals.org/D

ownloaded from

http://joeg.oxfordjournals.org/

In this article, we stand at the crossroads of these two streamlines of research andinvestigate the importance of firm and local technological relatedness—measured a laHidalgo et al. (2007)—in shaping the evolution of firms’ product space in Turkeyduring the period 2005–2009.

The introduction of new products represents a key factor for countries’ long-runprosperity. The continuous discovery of new production opportunities, indeed, carriesahead the process of creative destruction from which economic development takes off.Understanding how countries’ production structure evolves, though, entails looking atfirms’ innovative behaviour and at their knowledge accumulation process, which can beboth path- and place-dependent (Heimeriks and Boschma, 2014). We, therefore,explore whether a firm’s decision over new product addition is affected by cognitivecloseness between the new products, on the one side, and the existing firm own andlocal—NUTS 3 region—product baskets, on the other.

The investigation of this topic in the case of an emergent economy can be consideredof particular interest. On the one hand, product innovation in these economiesrepresents an opportunity to upgrade the production structure towards moresophisticated goods and, thus, to promote higher growth (Hausmann et al., 2007).On the other hand, compared with advanced economies, rapid and recent industrial-isation countries are usually characterised by initial important ancestral differences inthe geographical distribution of economic activities. They, therefore, constitute a goodempirical setting for studying the birth and evolution of clustering of industries acrossspace and firms, as the pace of industrialisation accelerates.

With this article, we aim at making three contributions: firstly, we focus on and assessthe impact of both firm and local product-specific capabilities on a firm’s decision overwhich products to add to its existing portfolio. For the first time to our knowledge, weexploit firm product-level data and inspect firms’ product choices, conditional on theirstatus of innovators. This is, in our view, the natural framework in which to analysepath and place dependence in the introduction of new products. As a matter of fact,differently from previous work on the role of within firm and local resources for theoverall firms’ innovation propensity, our empirical setting allows to identify theproduct-specific nature of firm and local competencies and to test its effectiveness inshaping the product space of firms. We, therefore, extend the country-level evidence onpath dependence in the productive structures’ evolution (David, 1985; Arthur, 1989;Hidalgo, 2009; Hausmann and Hidalgo, 2009, 2011) to the firm level and explain theemergence and persistence of geographically bounded product clusters that are at thebasis of long-lasting differences across regions.

Secondly, our work explores the role of several dimensions of firm heterogeneity interms of internal and environmental resources in relaxing the linkage between local andfirm pre-existing capabilities and the evolution path of production. Large, highlyproductive firms, firms engaged in R&D and in complex productions could betterexploit their internal resources to develop new capabilities and gain independence fromthe context they operate in. Foreign-owned firms, on their side, like exporters,importers and multi-plant firms located in different regions, operate within more openand wider networks and, thus, have weaker ties with the local environment(Granovetter, 1973). This can favour their escape from the historical regionalcomparative advantage, due to knowledge spillovers from their international networkand, for foreign firms, to the availability of larger intra-group financial resources (Desaiet al., 2004, 2008). In all these cases, regardless of local conditions, the availability of a

976 . Lo Turco and Maggioni




ownloaded from


larger knowledge base can importantly contribute to reduce a firm’s entry cost in abrand new technology (Perez and Soete, 1988).

Finally, to the best of our knowledge, this is the first piece of research investigatingthe importance of firm and local capabilities in shaping product innovation in Turkey.Relevant territorial disparities characterise the country. A laggard east contrasts with amore developed west, where most of the Turkish industrial base is located. In thiscontext, it is fundamental to ascertain whether and to what extent geographical andtechnological proximities explain the innovation performance of firms and the evolutionof the country’s product space. Our analysis, thus, contributes to assess the outcome ofrecent cluster policies that aim at favouring the spatial diffusion of local industrialdevelopment. Policies should indeed be directed to avoid that path- and place-dependent engender negative lock-in processes in regional development.

This article is organised as follows: Section 2 reviews the relevant literature; Section 3presents the sample, the data sources and our main relatedness indicators; Section 4describes the empirical strategy and presents the results; Section 5 concludes.

2. Geography, technological relatedness, and the relativecontribution of space and firm resources to product innovation

The literature has shown that knowledge externalities, which play a relevant role ineconomic growth (Arrow, 1962; Romer, 1986a,b; Grossman and Helpman, 1993), aregeographically localised (Jaffe et al., 1993) and importantly enhance firm innovation,especially in small- and medium-size firms (Audretsch and Feldman, 2004). Despite rapidspur and advances in information and communication technologies, the crucial differencebetween information and knowledge (Gertler, 2003; Howells, 2012) may still make thelocal environment matter in the process of innovation creation. Tacit knowledge, indeed,has an important role in developing successful firm routines and represents a keycompetitive element in a period of widespread access to codified knowledge (Maskell andMalmberg, 1999; Gertler, 2003). However, geographical proximity is neither a necessarynor a sufficient condition for spurring knowledge across firms (Boschma, 2005).On the onehand, as the innovation process increasingly benefits from learning by interacting (Lundvalland Johnson, 1994), the social dimension of innovation gains in importance and relationalproximity could substitute for the geographical one in fostering innovation. In this respect,the literature has shown that knowledge spillovers are localised, as long as the source ofknowledge (e.g. inventors) is geographically bounded (Breschi and Lissoni, 2001; Boggsand Rantisi, 2003; Breschi and Lissoni, 2005, 2009). On the other hand, other types ofproximitymay be necessary to complement the spatial one in favouring the introduction ofnew products. In order to learn from the local knowledge pool, firms have to absorb therelevant knowledge and, thus, need to be cognitively proximate to the local environment(Boschma, 2005). As a consequence, the notions of geographical and cognitive proximitiesbecome intimately related (Baptista and Swann, 1998; Orlando, 2000; Autant-Bernard,2001). More specifically, the concept of technological relatedness—meant as interactivelearning among economic actors stemming from the existence of a certain degree ofcognitive proximity among them (Timmermans and Boschma, 2014)—has, therefore,given new momentum to debate in the economic geography literature over the relativeimportance of regional variety versus specialisation in the mechanics of agglomerationexternalities for innovation and growth. In this line, Boschma et al. (2012), recently,

On firms’ product space evolution . 977




ownloaded from


investigate the importance of diversification in related industries for regional value-addedand employment growth in Spain at the NUTS 3 region level during the period 1995–2007.By comparing the cluster classification introduced by Porter (2003) and the proximityindex proposed by Hidalgo et al. (2007), they show that Spanish provinces with a widerange of related industries tend to enjoy higher economic growth rates. Focussing onregional competitiveness, Boschma et al. (2013) instead explore the role of regional- andcountry-level density measures around a product on a region’s probability to develop arevealed comparative advantage in that product. By using export data on 50 Spanishregions at the NUTS 3 level in the period 1988–2008, they show that proximity to theregional industrial structure plays a much larger role in the emergence of new comparativeadvantage industries in regions than does relatedness to the national industrial structure.This hints at important complementarities between geographical and cognitive proximityin the spur of knowledge, although the work shows that regional capabilities favour themaintenance rather than the development of comparative advantages. Neffke et al. (2011)analyse the economic evolution of 70 Swedish regions during the period 1969–2002 andfind strong path dependence in their diversification process. Their results confirm thattechnological relatedness—measured by a Revealed Relatedness indicator based on theratio between industries’ co-occurrences in regions’ industrial structures and their predictedvalue—is important in rising regions’ technological cohesion. Over the same period and forthe same country, Neffke et al. (2012) find that plant survival is importantly and positivelyaffected by technologically related localisation externalities. The role of relatednessbetween firms and the local pool of knowledge in Sweden is also confirmed by Boschma etal. (2009) who show the positive impact of intra-regional-related skill mobility on firms’productivity growth. Boschma and Iammarino (2009) also look at the role of ‘related’knowledge on regional economic growth in Italian provinces for the period 1995–2003.Their relatedness indicator hinges on the belonging of sectors/products to the same two-digit sector, thus following the notion proposed by Frenken et al. (2007). Differently fromother works, however, they highlight the important role of related extra-regionalknowledge in shaping the process of regional economic growth. In this line,Timmermans and Boschma (2014) find that plants’ productivity growth in Denmark isimportantly fostered by the inflowof skills that are cognitively close to existing skills and, inparticular, by inter-regional skilled labour mobility.

All the mentioned evidence suggests that a crucial role for the performance of economicentities is played by technological relatedness to the local or external pool of knowledge.

Technological proximity also matters within a firm’s boundaries and is found toaffect a firm’s diversification process. Firm growth, indeed, can be viewed as a processof exploitation of productive opportunities (Penrose, 1959). Firms are unique bundlesof resources where firm-specific abilities and general organisational routines arecombined with product-specific competencies related to the production of a particulargood.1 Thus, product-specific capabilities can constitute an important knowledge baseallowing firms to explore new production fields and diversify into technologicallyrelated products (Danneels, 2002). In this respect, on a sample of US, Italian, French,UK, German, and Japanese firms patenting to the European Patent Office from 1982 to1993, Breschi et al. (2003) show that knowledge relatedness, measured on the basis of

1 A similar view of the firm is reproduced by recent mainstream models of multi-product firms (Bernardet al., 2011).





ownloaded from


the co-classification codes contained in patent documents, is a major determinant offirm diversification, measured as a firm’s probability to be simultaneously active in afurther activity other than the core one. In a similar way, Neffke and Henning (2013)identify skill relatedness by using information on cross-industry labour flows and showthat firms are more likely to diversify into industries that are more ‘skill’ related to theircore activities than into industries without such ties or into industries that are linked byvalue chain linkages or by classification-based relatedness. These firm-level studies,thus, suggest that, although inter-firm technologically related knowledge spillovers doplay a role in the evolution of firms’ production structure, firms’ internal product-specific abilities are undoubtedly a key driver of their choice of new productions.

Actually, part of the literature on firm innovation has called into question theexistence of a regional environment effect tout-court. By comparing the relativeimportance of firm and local resources and capabilities, Pfirrmann (1994) for a sampleof Small and Medium Enterprises in Germany, Sternberg and Arndt (2001) on a sampleof European firms mainly of medium and small size and Beugelsdijk (2007) for a sampleof Dutch firms, all show that firm-specific resources are generally more important thana firm’s regional environment for any kind of product innovation. Although thisliterature casts a shadow on the relevance of inter-firm interactions within boundedterritories for firm growth and diversification processes, in our view, the concept oftechnological relatedness can redeem the absolute and relative contribution of localknowledge by shedding light on the circumstances under which and to what extent aregional environment can matter for product innovation.

In this article, we, therefore, test and compare the impact of firm and local product-specific capabilities on the process of firm introduction of new products in the contextof the Turkish emerging economy.

We expect both firm and local technological relatedness to drive a firm’s choice overnew product additions, nonetheless we believe that initial regional disparities could affectthe relative importance of these two factors. As evolutionary economic geographers haveargued, when starting brand new sectors/products, firms need to develop by themselvesthose product/sector-specific capabilities that cannot be retrieved in their location regions,which, instead, only provide general knowledge and skills (Boschma and Lambooy, 1999;Boschma and Frenken, 2006). This implies that firms’ own knowledge is fundamental inall those cases where a relevant gap exists between the new products’ requirements andthe local supply of product-specific factors. Hence, conditionally on observing newproduct additions, we expect firm capabilities to represent the main driver of the productspace evolution in Turkish laggard and peripheral regions. Owing to these areas’ tinyproduct base, firms are less likely to retrieve locally the necessary specific resources. Onthe contrary, the relative contribution of regional product-specific factors should behigher in developed regions, which are endowed with a larger and diversified productionstructure and, as a consequence, knowledge base.

As a further step of our analysis, we inspect the role of firm heterogeneity in affectingthe dependence of firms’ product portfolio choices on the firm internal and localproduct-specific capabilities. On the one hand, a minor role of the environment shouldbe recorded for large, internationalised, high Research & Development (R&D) intensiveand high complexity firms. On the other hand, we expect firms’ own knowledge toparticularly affect product additions of the less complex firms. The latter, indeed, aremore likely to introduce products whose knowledge base rests on a few competencies,which are already available within a firm’s own boundaries.





ownloaded from


3. Data and measurement issues

3.1. Data sources and estimation sample

Our firm product-level data set originates from the matching of the Turkish Annual

Industrial Product Statistics (AIPS), Structural Business Statistics (SBS) and Foreign

Trade Statistics (FTS).2 AIPS convey information that an all 10-digit PRODTR goods—

volume and value of production and sales—produced by Turkish firms with more than 20

persons are employed in either section C (mining and quarrying) or section D

(manufacturing) of NACE Rev 1.1 in the 2005–2009 time span. For firms in AIPS,

SBS provides information on firms’ output, input costs, employment, foreign ownership,

and NUTS 3 region of location. FTS reports then on firms’ import and export activities.The investigation of the role of both local and firm product-specific competencies called

for the construction of the relevant product space. While from AIPS, we could directly

observe the firms’ product mix, we retrieved the provincial product baskets by aggregating

data from firm to province level. In particular, in the computation of production value at

the province level, we had to cope with the presence of multi-plant firms in our database.

For single-product multi-plant firms, we assumed that the value of the single good

produced by each plant is proportional to its declared turnover. In order to split the

production of multi-product multi-plant firms across plants located in different NUTS 3

regions, we assumed, instead, that each plant produces the same products and we

attributed the production value to each plant in proportion to its turnover.3

Finally, in order to measure technological proximity and, thus, to compute the firm

and local product density indicators (Hidalgo et al., 2007), we use BACI (Gaulier and

Zignago, 2010) trade data at country-product level. BACI data are recorded according

to the 1996 HS classification. Production data, instead, are recorded according to the

PRODTR classification system, whose first six digits correspond to the CPA. In order

to match firm product-level production data with the proximity indicator computed on

the basis of product-level export information retrieved from BACI, we first converted

1996 HS flows into CPA by means of the HS-CPA correspondence table available from

RAMON website and we constructed a harmonised classification that is just slightly

more aggregated than the CPA classification, which we call HCPA. The latter contains

1297 products of which 1030 are actually produced in Turkey. Hereafter, product code

refers to HCPA classification.Our estimation sample is made up of all Turkish manufacturing and mining4 firms

with more than 20 persons employed introducing a new product in the 2006–2009 time

span. As we aim at highlighting how local and firm product technological relatedness

affect a firm’s decision over new product additions, our empirical results are conditional

on a firm’s product innovation status.

2 All sources are available from the Turkish National Statistical Office.3 We compared the territorial distribution of production stemming from this assumption to the emerging

one by attributing to the plant the production of those products falling within the four-digit NACE sectordeclared as the plant’s main activity. The two territorial distributions of production are rather close.Descriptive evidence is not shown for the sake of brevity and is available upon request.

4 The following empirical analysis is unchanged when the very few mining sector firms are removed fromthe sample. Results are not shown for the sake of brevity, nonetheless they are available from the authorsupon request.





ownloaded from


3.2. Measuring product innovation and product relatedness

3.2.1. Product innovation

To model a firm’s decision over which new good to produce, we need to define the set ofall possible product additions for each firm i. Allowing a firm to choose among any ofthe existing HCPA 1297 products not previously produced would make the analysiscomputationally unfeasible, due to the very high number—roughly 13 millions—offirm–product combinations. We, thus, restrict product addition possibilities to thoseproducts that, in a way, may be technically related according to a firm’s sector code(Frenken et al., 2007) and define the set of all possible product additions for each firm ias the set S2d including all products belonging to one of the two-digit NACE codeswhere the firm was actually active in the year preceding the product addition, that is attime t� 1. Our dependent variable Iip t is, therefore, a dummy taking value 1 for anyproduct p, firm i starts to produce at time t, and 0 for any product p 2 S2d that firm idoes not add to its product portfolio.

In our sample period, the share of new products with respect to all potential productsbelonging to one of the two-digit codes within which a firm was active in the previousyear is on average 1.78% and this percentage remains substantially unchanged duringthe period of our analysis. It follows that product addition is, indeed, a rare activity.

To assess the importance of new products for the geography of manufacturingproduction in Turkey, we analyse their impact on the evolution of industrial output acrossregions. In Figure A1 in the Appendix, Panel A documents the unequal distribution ofindustrial production between eastern and western regions at the beginning of our sampleperiod, while Panels B andC reveal that new products importantly contribute to explain thedynamics of regional industrial growth,5 especially in laggard regions. This implies that newproducts can represent an important factor helping to reduce the historical territorial divide.

3.2.2. Product relatedness

In order to calculate the extent of cognitive relatedness between firms’ new productsand their own and local capabilities, we hinge on the proximity indicator introduced byHidalgo et al. (2007) between each pair of products, which is based on the co-occurrence of products in the export basket of countries.6 We calculate this index as:

�pj ¼ minfPðRCAxpjRCAxjÞ;PðRCAxjjRCAxpÞg

that is, the distance between HCPA good p and HCPA good j is equal to the minimumbetween the probability that good p is exported conditional on good j being exported andthe probability that good j is exported conditional on good p being exported. TheHidalgo et al.’s (2007) proximity indicator, in our view, is a more comprehensive andsuperior proxy of technological and cognitive relatedness across products compared with

5 The evidence from Panel B is strongly supported by official data on the spatial distribution of averageannual export growth. As exports are out of the scope of our analysis, the corresponding map is notshown for brevity; nonetheless, it is available from the authors upon request.

6 Several further measures of product relatedness have been used in the literature (Teece et al., 1994; Fan andLang, 2000; Porter, 2003; Neffke and Henning, 2008; Bryce and Winter, 2009). However, the Hidalgo et al.’s(2007) proximity indicator is, in our view, the most comprehensive, detailed and less computationallydemanding. This is indeed confirmed by the widespread use of such an indicator in recent empirical work toinvestigate structural change issues (Poncet and de Waldemar, 2012; Felipe et al., 2012).





ownloaded from


the traditional sector/product classification. It is, indeed, able to capture the similaritybetween two goods in the use of capabilities, regardless of their location in the standardproduct classification. The lack of perfect overlapping with the product/sector definitionis shown in Table 1. Proximity decreases when considering more aggregate sector codes;however the mean and median proximities of goods belonging to the same four, three andtwo digits are quite close. Also, although the average proximity values between couples ofproducts belonging to different sectors are lower than within sectors, the maximum valuesrecorded within and across sectors are pretty similar, thus hinting at the fact that productproximity is a concept that goes beyond the conventional sector classification.

In order to measure the relatedness between firms’ and provinces’ capabilitiesendowment, on the one hand, and the new products firms introduce, on the other, weadopt the density measure proposed by Hidalgo et al. (2007). The latter measures theweight of links of the new product with a specific firm and local subset of products relativeto the product’s total proximity to all of the available products. Thus, it reflects the relativeproximity between the capabilities required for the new product and the main firm andlocal capabilities. For each firm i, the firm product density indicator is measured as follows:

densip ¼

XNj¼1

�pj � dij

XNj¼1

�pj

ð1Þ

where dij is a dummy equal to 1 when product j is produced by the firm and is equal to 0otherwise. This dummy variable, therefore, identifies the relevant subset of products asthose already produced by the firm.7

For each firm i, we calculate a measure of provincial density around product p as

denslip ¼XLi

l¼1

sil

XNj¼1

�pj � xljRCA

XNj¼1

�pj

2666664

3777775with

XLi

l¼1

sil ¼ 1 8i ð2Þ

In the formula, xljRCA is a dummy equal to 1 for products in which province l has acomparative advantage and equal to zero otherwise.8 In this case, the relevant subset of

7 We tried to adopt a further measure of firm density based on product p’s proximity to the firm’s mainproduct, but the analysis was basically unchanged. For the sake of brevity, we do not show these resultshere, nevertheless they are available upon request.

8 We compute the RCA index on the basis of the province product-level production data that we obtainedby aggregating plant-level production data at the province level for each HCPA code. Thus, province l hasa comparative advantage in product p if the following index is higher than 1:

RCApl ¼

yplPNj¼1 yjlPLl¼1 yplPL

l¼1

PNj¼1 yjl

where in the formula L is the total number of provinces in the Turkish economy, and y denotes the value ofproduction. The province RCA index is then computed by considering the whole Turkish territory asreference.





ownloaded from


products is made up of those products where the province has manifested its expertise

and, thus, owns a revealed comparative advantage. In the formula, the term in bracketsrepresents the local density of province l. To account for firms with plants located in

different provinces, sil indicates the weight of province l in firm i’s total turnover and Li

the total number of provinces where firm i is present with its own plants. Thus, for eachfirm, the local density around its new products is a firm-level variable that represents

the weighted average of province densities across all provinces where the firm has atleast one plant.

Our aim in the present article is, thus, to highlight whether technological proximity of

the firm and provincial production structure is a significant driver of firm’s choices over

new products and their abilities to expand their product basket. Preliminary evidenceon the positive relationship between technological relatedness and innovation is

available in Table 2 where t-tests reveal that newly introduced product–firm

combinations systematically present higher firm and local densities.

4. Empirical analysis

4.1. Empirical model and estimation issues

To explore whether the introduction of a new product is affected by its proximity to

firm and local existing capabilities, we estimate the following Linear Probability Model

(LPM):

Iip t ¼ a0 þ a1denslip t�1 þ a2densip t�1 þ �0Xi t�1 þ �i þ �p þ �t þ �ipt ð3Þ

where Iip t is the dummy identifying the introduction of new product p by firm i, defined

as above and our right-hand side variables of interest are denslip t�1 and densip t�1, which,

as previously stated, respectively measure local and firm density at time t – 1 around afirm’s potential new product p. Owing to the inclusion of the local and firm density

measures, our analysis is carried on the 2006–2009 panel. In the model, Xi t�1 is a vector

of firm-level characteristics all measured in t – 1 and including firm size, labourproductivity, export, import, foreign ownership status and a dummy for multi-plant

firms.

Table 1. Proximity values within group of HCPA codes

Mean Median SD Min Max

Across all products 0.172 0.160 0.104 0 0.864

Within the same two digits 0.223 0.212 0.119 0 0.864

Among different two digits 0.169 0.156 0.102 0 0.797

Within the same three digits 0.237 0.222 0.135 0 0.864

Among different three digits 0.171 0.159 0.103 0 0.797

Within the same four digits 0.245 0.231 0.142 0 0.864

Among different four digits 0.172 0.160 0.104 0 0.851

The authors’ elaborations on AIPS and SBS data and BACI dataset. The table shows the descriptive

statistics of the product proximity indicator � (Hidalgo et al., 2007) for different sub-samples of product

pairs, on the basis of their CPA sectoral classification.





ownloaded from


�i, �p and �t, respectively denote firm, product and time fixed effects, while �iptrepresents an idiosyncratic error term.

Table A1 in the Appendix contains a detailed description of all the right-hand sidevariables included in the various specifications of model 3.

Despite the pitfalls of the LPM, the latter does not need any distributionalassumption to model unobserved heterogeneity—in particular firm and product time-invariant characteristics that may drive a firm’s product choice—and in general deliversgood estimates of the partial effects on the response probability near the centre of thedistribution of the regressor (Wooldridge, 2002). As the LPM is affected byheteroskedasticity, our standard errors are robust and clustered by firm9 and ourpredicted probabilities always lie between 0 and 1. Nevertheless, in the robustnesschecks, we adopt alternative non-linear models to test the robustness of our findingsbased on the LPM.

4.2. Baseline results

Table 3 shows results for a baseline specification of model 3. In columns 1–6, only firmand local density measures are included in the specifications and estimates reveal that,regardless of the inclusion of year, product and firm fixed effects, both the existence offirm and local capabilities that are technologically proximate to those required for thenew product positively affect its introduction. The inclusion of fixed effects increasesthe coefficient of local density that becomes stable even when accounting for the impactof firm density. Thus, fixed effects reduce the interdependence between local and firmdensities around new products, by capturing those factors that are common to the firmand the local capabilities in determining the choice of new products. The role of firminternal product-specific capabilities is instead downsized by the inclusion of fixedeffects.

We control for further variables that are expected to affect a firm’s productinnovation activity: firm size and productivity levels, importer, exporter, foreignownership and multi-plant status dummies and the firm’s local RCA index. Detaileddescription of the computation of all the variables included in the analysis is available inTable A1 in the Appendix.

Table 2. The t-test

Iip¼ 1 Iip¼ 0 t-test

dens 0.007 0.004 �88.796

densl 0.287 0.254 �28.339

The authors’ elaborations on AIPS and SBS data and

BACI dataset. densl and dens captures local and firm

product relatedness measured by resting on the Hidalgo

et al.’s (2007) density indicator.

9 When we cluster standard errors at the product level, our results are not affected at all.





ownloaded from


Larger firms are more likely to invest in R&D and to gather the necessary financialresources, either internally or from the local banking system (Beck et al., 2005, 2008). Inaddition, for a given size, more efficient firms may more easily overcome the fixed costof innovating. Actually, recent literature on firm heterogeneity shows that firminnovation patterns are closely related to heterogeneous efficiency levels, which are, inturn, closely related to the diversity of firms’ export activities (Melitz and Burstein,forthcoming). Finally, firms active in international markets may be more likely toinnovate, as trade can be regarded as the flow of extra-regional knowledge (Boschmaand Iammarino, 2009). On the one hand, importers may access new, better quality andmore suitable inputs (Krishnan and Ulrich, 2001; Goldberg et al., 2009; Colantone andCrino, 2014). On the other hand, exporters may dramatically be pushed to innovate bytheir own foreign customers (Egan and Mody, 1992; Goh, 2005; Salomon and Shaver,2005; Baldwin and von Hippel, 2011; Hahn and Park, 2011; Bratti and Felice, 2012; LoTurco and Maggioni, 2015). Foreign-owned firms, then, besides being more export and

Table 3. Results: firm and local capabilities in product innovation

[1] [2] [3] [4] [5] [6] [7] [8] [9] [10]

densl 0.025*** 0.005*** 0.044*** 0.044*** 0.040*** 0.040***

[0.001] [0.001] [0.007] [0.007] [0.007] [0.008]

dens 2.706*** 2.657*** 1.860*** 1.860*** 1.848*** 1.846***

[0.071] [0.073] [0.190] [0.190] [0.205] [0.205]

Logdensl 0.005***

[0.001]

Logdens 0.008***

[0.001]

Size 0.002 0.001 0.001 0.001

[0.001] [0.001] [0.001] [0.001]

LP 0.000 0.000 0.000 0.000

[0.001] [0.001] [0.001] [0.001]

Importer �0.001 �0.001 �0.001 �0.001

[0.001] [0.001] [0.001] [0.001]

Exporter �0.001 �0.001 �0.001 �0.001

[0.001] [0.001] [0.001] [0.001]

Foreign 0.008** 0.008* 0.010** 0.009***

[0.003] [0.004] [0.004] [0.003]

Multi-plant 0.002 0.001 0.002 0.001

[0.001] [0.001] [0.001] [0.001]

RCAL 0.001*** 0.001*** 0.001*** 0.001***

[0.000] [0.000] [0.000] [0.000]

FE:

Year y y y y y y y y y y

Product n n n y y y y y y y

Firm n n n y y y y y y y

Observations 991,398 991,398 991,398 991,398 991,398 991,398 897,344 897,344 897,344 895,823

R2 0.001 0.008 0.008 0.079 0.079 0.079 0.081 0.082 0.082 0.082

�p < 0:10;��p < 0:05;��p < 0:01. Dependent variable: firm probability to introduce new product p.

Standard errors are clustered by firm. All regressors included in the estimation are lagged to one year.

densl and dens capture local and firm product relatedness measured by resting on the Hidalgo et al.’s (2007)

density indicator. Columns 7–9 control for further firm-level variables: firm size, Size, productivity, LP, a

dummy for the status of importer, Importer, a dummy for the status of exporter, Exporter, foreign

ownership dummy, Foreign, multi-plant dummy, Multi-plant, and the local RCA index, RCAl.

Column 10 tests for the logarithmic transformation of the density indicators.





ownloaded from


import intensive, may further benefit from technological spillovers from theirheadquarters and from the availability of intra-group financial resources (Desaiet al., 2004, 2008). The latter may offset the negative impact of financial constraints thatusually affect firms operating in less-developed economies (Gorodnichenko andSchnitzer, 2010). Besides this set of firm-level controls, we included a dummy variablefor multi-plant firms to address any possible externality stemming from theirsimultaneous presence in different locations. The inclusion of the local RCA indexvalue in the new product, RCAl, is aimed to ensure that the local density indicator doesnot actually capture the extent of local specialisation in that product (Hausmann andKlinger, 2007; Poncet and de Waldemar, 2012; Boschma et al., 2013) instead of theavailability of proximate capabilities around the firm’s local units. Results in columns7–9 show that, among firm-level characteristics, only foreign ownership exerts a positiveand significant direct effect on the introduction of new products. Among innovators,foreign firms are more likely to introduce a larger number of new products, possibly dueto their wide international network and their larger knowledge base. The poorperformance of other firm-level characteristics in predicting the pattern of productinnovation may stem from the inclusion of firm fixed effects in a short-panel dataset.Alternatively, it may also indicate that while firm characteristics determine firmpropensity to innovate, firm product-specific capabilities especially matter for thechoice of product additions. Furthermore, local specialisation in the product is asignificant determinant of the firm’s probability to introduce that product in thatlocation. However, the inclusion of this set of controls does not dramatically alter eitherthe size or significance or the relative importance of our density indicators.

In order to account for and compare the economic magnitude of the effects of localand firm densities around the new product, taking partial effects in column 9 as abenchmark and the average density values for innovating and non-innovating firmproduct groups from Table 2 above, we calculate the relative importance of firm andlocal resources for new product introductions. If the level of densities of non-innovatorsjumped to the values of innovators, a firm average innovation probability wouldincrease by 0.6 and by 0.1 percentage points because of firm and local density,respectively. These figures, respectively, imply an increase of 34% and 6% in theaverage innovation rate—1.78%—and hint at the possible higher responsiveness offirms’ new product choice to firm internal product-specific resources rather than to thelocal ones. In column 10 of Table 3, we actually confirm this insight. When we smooththe density indicators by taking their log, the estimated semi-elasticities imply a largerrole for firm rather than local product-specific capabilities. This finding is displayed inFigure A2 in the Appendix, which shows the relative importance of firm densitycompared with the total—firm and local—technological proximity in explaining thespatial distribution of firms’ introduction of new products.10 The comparison ofFigures A1 and A2 suggests that firm product-specific capabilities could have led thedynamics of industrial production and innovation experienced by laggard easternregions—such as Agri, Igdir, Hakkari, Van and Erzurum—in the time span consideredin our analysis. Local product-specific capabilities, instead, emerge as the main driver ofnew product introduction for all richer and more industrial regions, such as Ankara,

10 The relative importance of firm density in explaining NUTS 3 region average innovation rates iscalculated as a2�dens

a1�denslþa2�denswith coefficients �1 and �2 taken from Column 9 of Table 3.





ownloaded from


Istanbul, Manisa, Bursa and Konya. To further inspect whether the traditional Turkishterritorial divide somehow shapes the working of firm and local product-specificcapabilities, we ran the above analysis by sub-samples of western and eastern firm–product observations.11 Results are shown in column 1 of Table 4 and point at firm-level product-specific competencies having a statistically significant higher impact oneastern regions compared with the western ones. By the same token, when in column 2we focus on the split between regions with value—averaged over our sample time span,2006–2009—of new product introduction, Iip, above and below the median, again wefind that both local and, in particular, firm capabilities emerge as significant andrelevant elements in the catching up process of less-innovative areas. These findings arein line with a firm’s need to develop suitable internal resources and organisationroutines for the introduction of a new product, when a relevant gap exists between thelocal pool of competencies and those required by the new product. As a matter of fact,peripheral areas are characterised by a scant and poorly diversified set of economicactivities and the local environment can support product innovation to a very limitedextent. A firm’s own product-related knowledge and experience are instead fundamen-tal for its ability to innovate and, therefore, represent an important driver of structuralchange in the local product space.

4.3. Robustness checks

To check the robustness of the baseline findings, we run a number of checks.Firstly, we use alternative density indicators and results are shown in Table 5. In

column 1, firm local density—such as the local RCA measure—is based on RCAscalculated as the province product share over the world export share in the product,instead of on the Turkish production share in the good. This is to account for a possiblemis-measurement of provinces’ RCA indexes due to the use of Turkish industrialstructure as the benchmark for the comparison of each province production structure.In column 2, we enlarge the scope of local RCA products by defining a product withRCA when the latter indicator is above 0.5 rather than above 1. In column 3, to accountfor the possible dependence of our density measures on the number of products thatfirms produce and in which locations own a comparative advantage, we normalise themby dividing by the total number of firm products and the total number of local RCAproducts. For the same purpose of checking the robustness of our preferredtechnological proximity indicators to any possible scale effect, in column 4, wesubstitute average local and firm proximity for firm and local densities.12 In column 5,the local density indicator is referred to the NUTS 2 region(s) where the firm is active.All these changes, which try to account for potential pitfalls of the original densityindicators, leave our insights substantially unaltered.

Secondly, we allow for alternative sampling strategies and further firm and local levelcontrols in Table 6. In order to account for the possible selection bias in our estimates

11 Western NUTS 3 provinces are all the ones belonging to the following NUTS 1 regions: Istanbul, Bat iMarmara (Western Marmara), Ege (Egean region), Dogu Marmara (Eastern Marmara), Bat i Anadolu(Western Anatolia) and Akdeniz (Mediterranean region). The remaining provinces belong to the Eastarea.

12 Instead of computing a density indicator, we average the proximity indicator between the potentialproduct p, on the one hand, and each product produced by firm i and products in which the provinceenjoys an RCA, on the other hand.





ownloaded from


stemming from the exclusion of firm product observations for non-innovating firms, werandomly select 20% of them from the same two-digit codes where non-innovatingfirms in our sample are active and attribute to them the local and firm density measuresaround the products they could potentially introduce. Results are shown in column 1and are rather close to the baseline ones. The inclusion of non-innovators in the sampleallows for the identification of the impact of previous innovation activities on a firm’sprobability to add a specific product (Cefis and Orseningo, 2001; Cefis, 2003).

Table 4. Results: regional divide and catching up

[1] [2]

A B A B

Western regions Eastern regions Above P50 NUTS 3 average

of I ip Below P50

densL 0.037*** 0.088* 0.024*** 0.069*

[0.008] [0.051] [0.008] [0.036]

dens 1.756*** 4.007*** 1.766*** 3.025***

[0.208] [1.103] [0.219] [0.540]

Size 0.001 �0.006 0.001 �0.002

[0.001] [0.004] [0.002] [0.002]

LP 0 �0.001 0 �0.001

[0.001] [0.002] [0.001] [0.001]

Exporter �0.001 0.004 �0.001 �0.001

[0.001] [0.003] [0.002] [0.002]

Importer �0.001 �0.002 �0.002 0

[0.001] [0.004] [0.002] [0.002]

Foreign 0.010** 0 0.010** 0

[0.004] [0.000] [0.004] [0.000]

Multi-plant 0.002 0.001 0.002 �0.002

[0.001] [0.004] [0.002] [0.003]

RCAL 0.001*** 0.001*** 0.002*** 0.001***

[0.000] [0.000] [0.000] [0.000]

FE:

Year y y y y

Product y y y y

Firm y y y y

Observations 836,538 60,806 677856 219488

R2 0.084 0.109 0.088 0.078

Tests:

aA1 ¼ aB1 1.006 1.512

p-value 0.316 0.219

aA2 ¼ aB2 4.019 4.668

p-value 0.045 0.031

�p < 0:10;��p < 0:05;��p < 0:01. Dependent variable: firm probability to introduce a new product.

Standard errors are clustered by firm. All regressors included in the estimation are 1-year lags of the

corresponding variables.

Regressions are separately estimated for firms in groups A and B. The two groups are identified on

the basis of the following NUTS 3 characteristics: Western versus Eastern NUTS 3 regions, NUTS 3

regions above and below the median of the average introduction of new products.





ownloaded from


Table 5. Robustness: alternative density indicators

[1] [2] [3] [4] [5]

RCA based on world

product shares

Different RCA

threshold

Normalised

density

Proximity NUTS 2

Local unit

dens 1.852*** 1.849*** 1.844***

[0.206] [0.205] [0.205]

denslRCAWorld

0.016*

[0.009]

denslRCA05

0.020***

[0.006]

denslNorm 6.409**

[2.676]

densNorm 37.530***

[0.906]

�l 0.049***

[0.008]

� 0.153***

[0.004]

densR 0.045***

[0.008]

Size 0.001 0.001 0.002 0.001 0.001

[0.001] [0.001] [0.001] [0.001] [0.001]

LP 0 0 0 0 0

[0.001] [0.001] [0.001] [0.001] [0.001]

Importer �0.001 �0.001 �0.001 �0.001 �0.001

[0.001] [0.001] [0.001] [0.001] [0.001]

Exporter �0.001 �0.001 �0.001 �0.001 �0.001

[0.001] [0.001] [0.001] [0.001] [0.001]

Foreign 0.009** 0.010** 0.008*** 0.009*** 0.010**

[0.004] [0.004] [0.003] [0.003] [0.004]

Multi-plant 0.001 0.001 0.001 0.001 0.002

[0.001] [0.001] [0.001] [0.001] [0.001]

RCAl 0.001*** 0.001*** 0.001***

[0.000] [0.000] [0.000]

RCAlWorld

0.000***

[0.000]

RCAR 0.002***

[0.000]

FE:

Year y y y y y

Product y y y y y

Firm y y y y y

Observations 897,344 897,344 897,344 897,344 897,344

R2 0.081 0.082 0.089 0.088 0.082




Columns 1 and 2 test for the local density measures built by computing the local RCA indicator as the

provincial product share over the world export share and by selecting the local RCA products as the ones

displaying a RCA index above 0.5 rather than above 1, respectively. In Column 3, firm and local density

have been normalised by dividing them by the total number of firm products and the total number of local

RCA products. Column 4 explores the role of firm and local proximity indicators. In Column 5, the NUTS

3 local density indicator has been replaced by the more aggregated NUTS 2 local density indicator.





ownloaded from


Table 6. Robustness: further controls and firm-level determinants

[1] [2] [3] [4] [5] [6] [7]

Inclusion of

randomly

selected

20% of

non-innovators

Inclusion of

randomly

selected 20%

of non-innovators

and state

dependence in

innovation

Exclusion of

Istanbul

Drop of

provinces

with less

than 20

firms in AIPS

No. of

products in

firm and

province

Local

production

value

Further

firm

controls

densl 0.012*** 0.017*** 0.055** 0.042*** 0.077*** 0.030*** 0.037***

[0.003] [0.004] [0.025] [0.008] [0.010] [0.008] [0.008]

dens 1.037*** 1.296*** 2.327*** 1.861*** 3.703*** 1.851*** 2.287***

[0.142] [0.186] [0.419] [0.206] [0.202] [0.205] [0.204]

innot�1 0.003***

[0.001]

Size 0 �0.001 0.003 0.001 0.004*** 0.001 0.003*

[0.000] [0.001] [0.002] [0.001] [0.001] [0.001] [0.001]

LP 0 0.000 0 0 0 0 0

[0.000] [0.000] [0.001] [0.001] [0.001] [0.001] [0.001]

Importer 0 0.000 0 �0.001 0 �0.001 �0.001

[0.000] [0.001] [0.001] [0.001] [0.001] [0.001] [0.001]

Exporter 0 0.000 0.001 0 �0.001 �0.001 0

[0.000] [0.001] [0.001] [0.001] [0.001] [0.001] [0.001]

Foreign 0 0.002 0 0.010** 0.007 0.010** 0.013***

[0.002] [0.002] [0.000] [0.004] [0.005] [0.004] [0.005]

Multi-plant 0 0.000 0.001 0.002 0.002* 0.001 0.002

[0.000] [0.001] [0.002] [0.001] [0.001] [0.001] [0.001]

RCAL 0.001*** 0.001*** 0.001*** 0.001*** 0.001*** 0.001*** 0.001***

[0.000] [0.000] [0.000] [0.000] [0.000] [0.000] [0.000]

N �0.031***

[0.001]

NRCAL �0.011***

[0.002]

yl 0.000***

[0.000]

R&D 0.027**

[0.013]

w �0.001

[0.002]

Multi-product �0.019***

[0.001]

FE:

Year y y y y y y y

Product y y y y y y y

Firm y y y y y y y

Observations 1,642,911 1,046,698 415,209 891,101 897,344 897,344 897,344

R2 0.059 0.062 0.076 0.082 0.085 0.082 0.083




In Column 1, a randomly selected sample of 20% of innovators is included in the estimation. In Column 2,

we add a dummy for product innovators at time t – 1. In Columns 3 and 4, firms with at least one plant

located in the Istanbul province and firms located in all the provinces for which AIPS presents less than 20

firms have been excluded from the analysis, respectively. Column 5 controls for the number of products

produced by the firm, N, and the number of local RCA products, NRCAL. Column 6 controls for the local

value of production in the product, yl. Column 7 finally adds further firm-level variables: the share of R&D

workers, R&D, the average wage, w, and a dummy for multi-product firms, Multi-product.





ownloaded from


Therefore, in column 2, we account for state dependence in innovation by including adummy variable, innot�1 taking value 1 for firms that introduced a new product in year

t – 1 and zero otherwise.13 Accounting for past innovation compelled us to exclude year

2006, thus retaining in the sample only those firms for which we have the informationon production data for at least 3 consecutive years. Nonetheless, results on the

significance and relative importance of the two density measures are unaffected. Incolumn 3, we exclude those firms that have at least one plant located in the Istanbul

province from the analysis. This is to verify that our findings are not just driven by this

province, which presents a higher agglomeration of firms and variety of products. Incolumn 4, to check the validity of our provincial production data aggregated from firm-

level information, we exclude those provinces for which we have less than 20 firms in

AIPS. The small number of firms used to reconstruct the production data may indeedlead to poor aggregate production value proxies.

Furthermore, in order to ensure that our firm and local density indicators are not

capturing any generic effect of agglomeration economies, rather than the availability ofcognitively proximate firm and local capabilities, we include, in column 5, the log of the

number of products produced by the firm and the log of the number of products inwhich its province(s) has a revealed comparative advantage.14 Also, in column 6, we

include the local value of production in the product. On the one hand, the resulting

findings imply that Jacobs diversification economies positively operate only iftechnologically related. Indeed, a higher number of local RCA bears a negative and

significant coefficient. On the other hand, although a higher local production of the

good has a positive impact on a firm’s probability to introduce it, local density aroundthe product still matters, thus witnessing that a large existing local production scale is

not a necessary condition to develop new products in a region. We include furtherobservable firm characteristics at our disposal to check that our evidence is not driven

by the omission of other important time-varying firm-level characteristics possibly

driving a firm’s choice of products to add to its product basket. We, thus, introduce theshare of R&D workers, the average wage and a dummy for multi-product firms. The

latter are found to have a lower propensity to innovate, whereas a higher share of R&Dworkers is positively related to a larger extent of product innovation. Our insights are

unchanged in all cases.Thirdly, we tested the robustness of our findings to alternative sample compositions

and modelling choices and results are shown in Tables A2 and A3 in the Appendix.

The first table shows findings stemming from different sample selection rules.

13 To further control for both sample selection and state dependence in innovation activity, we haveimplemented a two-step empirical analysis. In the first step we estimated, by means of system GMM, adynamic linear probability model of a firm’s probability to introduce a new product. We, therefore,included as explanatory variables, the lagged innovation status and a large set of firm-level covariates.From this first step analysis where we specifically control for state dependence, we retrieved the residualand, in a second step, we include it as regressor in the estimations of the probability to introduce aspecific product p on the sample of innovators. The first-step regression reveals the importance of statedependence. Also, the residual is statistically significant, thus proving the existence of a sample selectionbias, and bears a positive coefficient, thus revealing that innovators at time t� 1 are more likely tointroduce a new product p at time t. Our findings on the role of firm and local product-specificcapabilities stay unchanged. Results are not shown for the sake of brevity, nonetheless they are readilyavailable from the authors upon request.

14 For firms located in different provinces, we consider the weighted average number of RCA productsacross the provinces of localisation.





ownloaded from


In columns 1–3, we differently select our sample within our basic conservative definitionof new product, where zero values are attributed to those HSCPA products that are notintroduced by the firm and belong to any of the two digits where the firm waspreviously active. First, we repeat our baseline estimation on a subsample obtained by arandom draw of 10% of the zero values in order to take into account the low percentageof 1 we have (column 1).15 Second, we discard those products whose production wasactually initiated, but whose NACE code falls outside any of the firm’s two-digit NACEcode (column 2), thus having a homogeneous definition of both the 1 and 0 values.Third, we focus on the sample of new potential products belonging to any two-digitNACE code where the firm was active, but which do not belong to any four-digitNACE code where the firm was already producing (column 3). In all these cases, weadopt a narrower diversification definition and, nevertheless, our main findings areunchanged. In columns 4–8, instead, we relax the definition of diversificationpossibilities and expand the number of possible product choices by considering all ofthe 1297 possible HCPA products present in our classification as new potentialproducts. In this respect, we also include those goods belonging to those two-digit codeswhere the firm was not active in t� 1 in firms’ addition possibilities set (Neffke andHenning, 2013). Thereby, we consider a more radical definition of innovation.However, this choice substantially increases the number of potential firm–productcombinations to include in our analysis. Owing to computational constraints, weimplement this analysis by year and by selecting the 10% of zero combinations. Finally,in Table A3 in the Appendix, we present estimates of non-linear models for the firm’sproduct choice. As running a conditional logit for the whole sample was computa-tionally unfeasible, we present estimates of a conditional logit model by year in columns1–4. In columns 5–7, we use logit, rare event logit16 and probit models.

All of this further robustness checks confirm the relevance of firm and localtechnological relatedness for the introduction of new products.

4.4. Does firm heterogeneity shape the role of capabilities?

Results from the tables mentioned above show the importance of the existing firm andlocal capabilities as a driver of firms’ product space evolution. However, they alsoreveal that firm-level heterogeneity hardly directly affects the choice of which productto introduce. This can be due to the introduction of fixed effects in the model with theshort time span at our disposal. However, a further possibility is that, rather thanexerting a direct effect, firm heterogeneity dimensions act as mediating factors of localand firm product-specific capabilities absorption. The existing set of local capabilitiesaround a product could play a heterogeneous role in orienting the firms’ productionchoices according to different endowment levels of firm internal resources andaccording to the extent of diversity of the firm-specific environment. On the one hand,as larger, more productive, R&D-intensive firms and firms producing moresophisticated goods own an important stock of knowledge on their own, their extentof innovation could be less affected by the local pool of product-specific capabilities.They could, then, be more autonomous in their innovation efforts than smaller, less

15 We also tested for different shares of zero observations.16 The ReLogit STATA command has been used.





ownloaded from


productive and less R&D-intensive firms and firms producing simple goods. On theother hand, internationalised firms such as multi-plant firms are active in a largernumber of diverse domestic, international and more knowledge-intensive networks (i.e.foreign networks are, indeed, made up of more competitive and productive firms) and,as a consequence, can draw on several pools of different knowledge stocks. Theirexposure to multiple environments reduces the contribution of product-specificcapabilities available in a particular geographical location to their innovation effort.

In a dynamic capabilities perspective, all these firm internal and environmentalfeatures allow firms to better grasp and exploit new opportunities across availabletechnologies and existing markets (Teece, 2007). Thus, we repeat the estimation inEquation (3) for different groups of firms and we compare small versus large firms,high- versus low-productivity firms, high- versus low-product sophistication firms andhigh versus low R&D employment share firms. Product sophistication is measured a laHausmann et al. (2007) by means of the Prody indicator, which measures the incomecontent of the product. Finally, to account for heterogeneous environmental exposureof firms, we compare multi- versus single plant firms, exporters versus non-exporters,regular exporters versus non-regular exporters, importers versus non-importers andforeign-owned versus domestic firms.

Results are shown in Tables 7 and 8. The last four rows of these tables present theWald test for equality of the density coefficients of the two groups that are compared,group A and group B.17 Table 7 focusses on heterogeneous effects according to differentlevels of some specific internal firm-level characteristics. From the estimations, itemerges that the local set of product-specific capabilities play a major role for theproduct introduction in smaller than in larger firms. Among the remaining firm groups,no statistically significant difference emerges in the local density coefficient estimates.Interestingly enough, firm density is statistically higher for firms producing less complexproducts, thus disclosing that, when a firm is endowed with a small set of capabilities, asrevealed by its simple productions, its flexibility in production shrinks, and the product-specific knowledge acquires a fundamental role for the innovation. The coefficientsestimates for high and low R&D employment share firms would imply a higher returnfrom internal resources for high R&D firms and a higher return from local resources forthe low R&D ones. These results are in line with the higher ability of high R&D firms toinnovate on the basis of their own skills and the greater low R&D firms’ need of thelocal pool of capabilities to introduce new products. However, possibly due to the smallnumber of firms hiring R&D workers in our sample, the differences between the twosets of coefficients are not statistically significant.

In Table 8, instead, we inspect the importance of the firm’s business network inmagnifying/reducing the impact of firm and local technological relatedness. We findthat the coefficient on local density is statistically significantly lower for multi-plant,exporting and importing firms. As expected, these firms are less affected by the localpool of capabilities when introducing a new product. From Table 8, an interestingfinding concerns regular exporters. Our data, indeed, allows for the identification ofexported products that are actually produced by the firm and exported products that,

17 When the variable of interest is not a dichotomous one, the two groups are defined as above and belowthe median of the corresponding indicator. As an example, large and small firms are those whose labourunits are above=equal or below the median of the variable size that identifies the number of employedpersons in the firm.





ownloaded from


instead, correspond to a pure trade intermediary activity of the firm. Regular exportersare firms that export at least one of their own produced goods. Estimates imply a higherreturn from firm capabilities in terms of innovation for this group of firms comparedwith non-regular exporters. This result could suggest that a deep and direct linkagebetween the firm own production and export activity enhances the acquisition andaccumulation of product-specific knowledge that can be exploited to innovate. Finally,no statistically significant difference emerges between foreign and domestic firms,although the coefficients of the two density measures are significant only on thedomestic firms’ sub-sample. This outcome, however, could be driven by the very smallnumber of foreign-owned firms in our sample.

Table 7. Results: heterogeneous internal resources

[1] [2] [3] [4]

A B A B A B A B

Large Small High

Productivity

Low

Productivity

High

Prody

Low

Prody

R&D No R&D

densl 0.030*** 0.066*** 0.037*** 0.065*** 0.046*** 0.042*** 0.031* 0.045***

[0.011] [0.014] [0.011] [0.016] [0.012] [0.012] [0.019] [0.009]

dens 2.357*** 2.089*** 2.150*** 2.594*** 0.706** 2.087*** 2.886*** 1.909***

[0.279] [0.288] [0.310] [0.295] [0.306] [0.239] [0.817] [0.215]

Size �0.001 0.001 �0.002 0.001 0.001 0.001

[0.002] [0.002] [0.001] [0.002] [0.004] [0.002]

LP 0 �0.002* �0.001 0 �0.001 0

[0.001] [0.001] [0.001] [0.001] [0.003] [0.001]

Exporter �0.003 �0.001 �0.003 0.002 �0.001 �0.001 �0.012* 0

[0.003] [0.002] [0.003] [0.002] [0.001] [0.002] [0.006] [0.001]

Importer �0.004 0.002 �0.003 0 �0.001 �0.002 0.005 �0.001

[0.003] [0.002] [0.002] [0.002] [0.001] [0.002] [0.004] [0.001]

Foreign 0.017** 0.002 0.013 0.002 0.008* 0.009 0.013 0.011*

[0.008] [0.007] [0.009] [0.005] [0.005] [0.009] [0.013] [0.006]

Multi-plant 0 0.003 �0.001 0.003 0 0.004 �0.004 0.002*

[0.002] [0.002] [0.002] [0.002] [0.001] [0.002] [0.006] [0.001]

RCAl 0.002*** 0.001*** 0.001*** 0.001*** 0.001*** 0.002*** 0.001*** 0.001***

[0.000] [0.000] [0.000] [0.000] [0.000] [0.000] [0.000] [0.000]

FE:

Year y y y y y y y y

Product y y y y y y y y

Firm y y y y y y y y

Observations 438,368 458,976 448,340 449,004 447,746 449,598 83184 814160

R2 0.088 0.084 0.078 0.095 0.069 0.098 0.09 0.084

Tests:

aA1 ¼ aB1 4.036 0.049 0.391 0.46

p-value 0.045 0.144 0.825 0.498

aA2 ¼ aB2 0.447 12.62 27.856 1.335

p-value 0.504 0.3 0 0.248




Regressions are separately estimated for firms in groups A and B. The two groups are identified as above

and below the median values of firm size, productivity, firm Prody and R&D employment share. The value

of the t-test for the difference between the coefficients of the two groups and the their p-value are reported

at the bottom of the table.





ownloaded from


4.4.1. Contribution of local and firm product-specific resources to innovationrates by firm groupings

Table 9 shows back of the envelope calculations to assess what would happen to

innovation rates in each group of firms if firm and local densities for non-innovating

firm–product pairs jumped to the values observed for innovating firm–product pairs.18

We take the observed differences of densities between observations with Iip¼ 1 and

observations with Iip¼ 0, multiply them by the estimated coefficients on local and firm

densities, respectively, and divide them by the group’s observed innovation rate, Iip%.

Thus, we get how much of the observed innovation rate is explained by firm and local

product-specific capabilities.

Table 8. Results: heterogeneous environmental resources

[1] [2] [3] [4] [5]

A B A B A B A B A B

Multi-

plant

Single-

plant

Exporters Non-

exporters

Regular

exporters

Non-

regular exporters

Importers Non-

importers

Foreign Domestic

densl 0.027** 0.090*** 0.035*** 0.082*** 0.034*** 0.053*** 0.027*** 0.114*** 0.038 0.039***

[0.013] [0.013] [0.010] [0.016] [0.011] [0.011] [0.008] [0.018] [0.025] [0.008]

dens 2.216*** 2.026*** 2.116*** 2.105*** 2.971*** 1.973*** 2.072*** 2.047*** 0.629 1.896***

[0.370] [0.256] [0.263] [0.394] [0.312] [0.297] [0.231] [0.532] [1.439] [0.203]

Size 0.003 0 0 0.001 �0.004 0.002 0.001 0.003 0.002 0.001

[0.003] [0.002] [0.002] [0.003] [0.003] [0.002] [0.002] [0.003] [0.008] [0.001]

LP 0 �0.001 0 0.002 �0.001 0 0 0 0.003 0

[0.001] [0.001] [0.001] [0.002] [0.001] [0.001] [0.001] [0.002] [0.004] [0.001]

Exporter �0.002 0 �0.004* 0.002 �0.015* 0

[0.003] [0.002] [0.002] [0.002] [0.008] [0.001]

Importer �0.002 �0.001 �0.001 �0.004** �0.001 �0.003** 0.012*** �0.001

[0.003] [0.002] [0.002] [0.002] [0.003] [0.002] [0.003] [0.001]

Foreign 0.007 0.015** 0.012** 0 0.011* 0.016*** 0.012** 0.014**

[0.006] [0.006] [0.006] [0.000] [0.006] [0.003] [0.005] [0.006]

Multi – plant 0.001 0.006** 0.001 0.003 �0.001 0.004** �0.002 0.002

[0.002] [0.003] [0.003] [0.002] [0.002] [0.002] [0.006] [0.001]

RCAl 0.001*** 0.001*** 0.001*** 0.001*** 0.001*** 0.001*** 0.002*** 0.001*** 0 0.001***

[0.000] [0.000] [0.000] [0.000] [0.000] [0.000] [0.000] [0.000] [0.000] [0.000]

FE:

Year y y y y y y y y y y

Product y y y y y y y y y y

Firm y y y y y y y y y y

Observations 301,654 595,690 559,377 337,967 363,374 533,970 573,304 324,040 27,047 870,297

R2 0.085 0.085 0.082 0.095 0.087 0.088 0.083 0.095 0.141 0.083

Tests:

aA1 ¼ aB1 11.397 6.252 1.472 18.803 0.002

p-value 0.001 0.012 0.225 0 0.962

aA2 ¼ aB2 0.177 0 5.368 0.002 0.76

p-value 0.674 0.983 0.021 0.966 0.383




Regressions are separately estimated for firms in groups A and B. The two groups are identified on the basis

of the following firm status: multi-plant firm, exporter, regular exporter, importer and foreign-owned firm.

18 For convenience of exposition, we omit calculations for the subgroups of domestic and foreign firms, asno significant difference actually emerged in Table 8.





ownloaded from


In line with the previous results concerning different levels of firm internal resources,a higher relative importance of local capabilities for smaller firms emerges. In addition,we find that the contribution of the local environment is much higher, both in absoluteand relative terms, for less-productive and non-R&D-intensive firms, regardless of anystatistically significant difference in the coefficient estimates. The contribution of thelocal set of product-specific capabilities is much higher for less complex products inabsolute terms, although only local resources appear to drive the innovation rate ofmore complex firm–product pairs.

Turning to firm heterogeneity based on firms’ exposure to more or less diverseenvironments, the Table shows that the local environment particularly affectsinnovation rates of single plant and non-internationalised firms, while firm internalresources, both in absolute and relative terms, have a much more relevant role forregular exporters and for importers especially. Thus, we do find some differences in theobserved relative contribution of firm internal resources for large, R&D-intensive,highly productive, multi-plant and internationalised firms, regardless of the lack of astatistically significant difference in coefficient estimates.

5. Conclusions

With this article, we have contributed to the literature on the relevance of technologicalrelatedness in shaping spatial clusters of economic activities and to the debate over therelative importance of firm and local resources for product innovation. For the firsttime to our knowledge, we have estimated a model of the impact of firm and local

Table 9. Contribution of local and firm capabilities by firm groupings

Large Small High

productivity

Low

productivity

High

Prody

Low

Prody

R&D No

R&D

denslðIip ¼ 1Þ � denslðIip ¼ 0Þ 0.032 0.035 0.035 0.035 0.006 0.046 0.016 0.002

densðIip ¼ 1Þ � densðIip ¼ 0Þ 0.003 0.002 0.003 0.003 0.00004 0.004 0.035 0.003

OBSERVED INNOVATION RATE (%)

Iip% 1.9 1.7 1.7 1.9 1.25 2.38 1.41 1.82

CONTRIBUTION OF LOCAL AND FIRM CAPABILITIES TO THE OBSERVED INNOVATION RATE (%)

a1 �densl ðIip¼1Þ�dens

l ðIip¼0Þ

Iip% 5.1 13.6 7.7 11.9 2.24 8.17 3.46 8.60

a2 �densðIip¼1Þ�densðIip¼0Þ

Iip% 42.9 29.0 39.6 39.7 0.21 35 42.55 31.27

Multi-

plant

Single-

plant

Exporters Non-

exporters

Regular

exporters

Non-

regular

exporters

Importers Non

importers

denslðIip ¼ 1Þ � denslðIip ¼ 0Þ 0.034 0.032 0.031 0.037 0.036 0.032 0.036 0.032

densðIip ¼ 1Þ � densðIip ¼ 0Þ 0.032 0.003 0.003 0.003 0.004 0.002 0.004 0.002

OBSERVED INNOVATION RATE (%)

Iip% 1.83 1.75 1.75 1.83 1.72 1.82 1.82 1.72

CONTRIBUTION OF LOCAL AND FIRM CAPABILITIES TO THE OBSERVED INNOVATION RATE (%)

a1 �densl ðIip¼1Þ�dens

l ðIip¼0Þ

Iip% 5.1 16.6 6.2 16.6 7.0 9.4 5.0 19.6

a2 �densðIip¼1Þ�densðIip¼0Þ

Iip% 35.7 34.2 36.4 31.9 64.3 26.4 41.0 19.4





ownloaded from


product-specific capabilities on a firm’s choice over new products by exploiting a firm

product-level sample of innovators from the Turkish economy. We have measured

technological relatedness by means of Hidalgo et al.’s (2007) density indicator that

defines two products as related on the basis of their co-occurrence in countries’ export

baskets. Our empirical analysis has conveyed three main findings.First, a firm’s activity of new product introduction shows strong path and place

dependence that originates from the availability of firm and local product-specific

competencies, cognitively proximate to those needed for the new goods’ production.

This result holds once accounted for a large number of robustness checks and, in

particular, for the effect of local Jacobs and specialisation external economies. This

implies two major facts. On the one hand, regional competencies matter for firm

innovation when they are cognitively close to the new products. We, thus, corroborate

previous evidence which discloses the positive role of existing local diversification in

technologically related sectors on the economic performance of regions and firms

(Boschma et al., 2009; Neffke et al., 2011; Boschma et al., 2012, 2013). On the other

hand, a large existing local production scale is not a necessary condition to develop new

products in a region.Secondly, new product choices are more responsive to firm internal resources than to

the local set of available product-specific knowledge. Also, the former represent the

main thrust of innovation in laggard and less-innovative areas, due to the relevant gap

between the existing scant and poorly diversified local capabilities and the ones needed

for the new products. Firms located in these areas, therefore, appear as autonomously

developing and adapting their specific abilities and general organisational routines to

the new products’ requirements. On the contrary, in more advanced regions, the

introduction of new products is sustained relatively more by the local availability of a

wide and thick pool of diverse technologically related knowledge.Third, firm heterogeneity acts as a mediating factor of the impact of firm and local

product-specific capabilities on product innovation. Firm size and international

exposure play a role in relaxing ties with the local environment for a given level of

firm internal product-specific resources, while a firm’s regular export activity and low

production complexity enhance the effect of firm product competencies for a given level

of local product-specific capabilities.Despite the minor role of space vis-a-vis the key role of firm knowledge for product

innovation in Turkish peripheral areas, an accurate interpretation of the overall

evidence above casts a shadow over the possibility that entrepreneurs could, by

themselves, restore and redeem the future of entire territories. As a matter of fact, what

matters for development and long-run growth is not the rate of innovation per se, but

the level of new products’ sophistication (Hausmann et al., 2007). In this respect, our

findings show that internal resources mainly explain the innovation rate of low

complexity firms. Hence, path dependence could doom firms and their location regions

to be locked in low-complexity product specialisation and low growth. Product-space

sophistication is a challenging task for a single firm, since the introduction of complex

goods importantly originates from the accumulation of a pool of technologically related

diverse scientific knowledge (Hausmann and Hidalgo, 2009; Heimeriks and Boschma,

2014). Therefore, the required capabilities, when missing, need to be sourced externally.

Our work, thereby, contributes to the debate calling into question the existence of a

regional environment effect on innovation: by accounting for the product-specific





ownloaded from


nature of skills and competencies, we, in fact, suggest that the local product space is ofprimary relevance for the introduction of new complex goods.

Findings in this article suggest that, in order to promote firms’ diversification andgrowth, the process of knowledge creation and acquisition should be enhanced.Therefore, our work conveys two important policy implications for Turkey and foremerging economies.

On the one hand, the evidence on spatial heterogeneity in path and place dependencethat characterises the uneven Turkish economic geography calls for crucial actions infavour of laggard regions directed to improve their endowment of generic resources andof factors that are specific to targeted sectors. Investments in infrastructures and humancapital would foster the creation of a favourable environment for entrepreneurialactivities. Reducing communication costs and improving the quality of local workforcewould also enhance the provision of efficient business services that are essential for theperformance of manufacturing firms. Nonetheless, to promote diversification in non-traditional sectors, policies could enhance the role of universities as centres of specificknowledge creation and incubators of spin-off firms. In this respect, ongoing support bythe Turkish government in favour of cluster policies around traditional low complexitygoods (e.g. carpets and wood products) should be complemented by actions directed tothe acquisition of specific factors to encourage laggard regions’ transition towardssmart specialisation products.

On the other hand, public interventions in support of firms’ international activitieswould ease their access to extra-regional knowledge pools.

All of these policy efforts would activate a self-reinforcing process of knowledgeproduction that would allow firms and territories to relax the dependence on historicallocal specialisation and co-evolve towards a larger and more sophisticated industrialbase.

These concluding remarks raise some new questions that still remain unansweredand, yet, are crucial to better understand the role of path and place dependence indevelopment. Is the extent of local complexity—meant as the accumulation of a rangeof diverse product-specific tacit and codified knowledge—a direct driver of a firm’sproduct mix sophistication? Is technological relatedness a mediating factor of such alinkage? Does a firm’s embeddedness in the local environment foster the sophisticationof its new goods? These issues constitute a relevant future research agenda.

Disclaimer

The data used in this article are from the Foreign Trade Data, the Annual BusinessStatistics and the Production Surveys provided by the Turkish Statistical Office(Turkstat). All elaborations have been conducted at the Microdata Research Centre ofTurkstat under the respect of the law on the statistic secret and the personal dataprotection.

Acknowledgement

We are grateful to TurkStat staff from foreign trade statistics and dissemination departments fortheir help. The results and the opinions expressed in this article are exclusive responsibility of theauthors and, by no means, represent official statistics.





ownloaded from


References

Arrow, K. (1962) The economic implications of learning by doing. Review of Economic Studies,29: 155–173.

Arthur, W. B. (1989) Competing technologies, increasing returns, and lock-in by historical events.The Economic Journal, 99: 116–131.

Audretsch, D. B., Feldman, M. P. (2004) Knowledge spillovers and the geography of innovation(chapter 61). In J. V. Henderson and J. F. Thisse (eds) Handbook of Regional and UrbanEconomics, vol 4, pp. 2713–2739. North-Holland, Amsterdam: Elsevier.

Autant-Bernard, C. (2001) The geography of knowledge spillovers and technological proximity.Economics of Innovation and New Technology, 10: 237–254.

Baldwin, C., von Hippel, E. (2011) Modeling a paradigm shift: From producer innovation to userand open collaborative innovation. Organization Science, 22: 1399–1417.

Baptista, R., Swann, P. (1998) Do firms in clusters innovate more? Research Policy, 27: 525–540.Beck, T., Demirguc-Kunt, A., Laeven, L., Levine, R. (2008) Finance, firm size, and growth.Journal of Money, Credit and Banking, 40: 1379–1405.

Beck, T., Demirguc-Kunt, A., Maksimovic, V. (2005) Financial and legal constraints to growth:does firm size matter? Journal of Finance, 60: 137–177.

Bernard, A., Redding, S., Schott, P. (2011) Multi-product firms and trade liberalization. TheQuarterly Journal of Economics, 126: 1271–1318.

Beugelsdijk, S. (2007) The regional environment and a firm’s innovative performance: A plea fora multilevel interactionist approach. Economic Geography, 83: 181–199.

Boggs, J. S., Rantisi, N. (2003) The ‘‘relational turn’’ in economic geography. Journal ofEconomic Geography, 3: 109–116.

Boschma, R. (2005) Proximity and innovation: a critical assessment. Regional Studies, 39: 61–74.Boschma, R., Eriksson, R., Lindgren, U. (2009) How does labour mobility affect the performanceof plants? The importance of relatedness and geographical proximity. Journal of EconomicGeography, 9: 169–190.

Boschma, R., Frenken, K. (2006) Why is economic geography not an evolutionary science?Towards an evolutionary economic geography. Journal of Economic Geography, 6: 273–302.

Boschma, R., Iammarino, S. (2009) Related varieties, trade linkages, and regional growth in Italy.Economic Geography, 85: 289–311.

Boschma, R., Lambooy, J. G. (1999) Evolutionary economics and economic geography. Journalof Evolutionary Economics, 9: 411–429.

Boschma, R., Minondo, A., Navarro, M. (2012) Related variety and regional growth in Spain.Papers in Regional Science, 91: 241–256.

Boschma, R., Minondo, A., Navarro, M. (2013) The emergence of new industries at the regionallevel in Spain: a proximity approach based on product relatedness. Economic Geography, 89:29–51.

Bratti, M., Felice, G. (2012) Are exporters more likely to introduce product innovations? TheWorld Economy, 35: 1559–1598.

Breschi, S., Lissoni, F. (2001) Localised knowledge spillovers vs. innovative milieux: knowledge‘tacitness’ reconsidered. Papers in Regional Science, 80: 255–273.

Breschi, S., Lissoni, F. (2005) ‘‘cross-firm’’ inventors and social networks: localized knowledgespillovers revisited. Annales d’Economie et de Statistique, 79/80: 189–209.

Breschi, S., Lissoni, F. (2009) Mobility of skilled workers and co-invention networks: an anatomyof localized knowledge flows. Journal of Economic Geography, 9: 439–468.

Breschi, S., Lissoni, F., Malerba, F. (2003) Knowledge-relatedness in firm technologicaldiversification. Research Policy, 32: 69–87.

Bryce, D., Winter, S. (2009) A general inter-industry relatedness index. Management Science, 55:1570–1585.

Cefis, E. (2003) Is there persistence of innovative activities? International Journal of IndustrialOrganization, 21: 489–515.

Cefis, E., Orseningo, L. (2001) The persistence of innovative activities. A cross-country and cross-sectors comparative analysis. Research Policy, 30: 1139–1158.

Colantone, I., Crino, R. (2014) New imported inputs, new domestic products. Journal ofInternational Economics, 92: 147–165.





ownloaded from


Danneels, E. (2002) The dynamics of product innovation and firm competences. StrategicManagement Journal, 23: 1095–1121.

David, P. (1985) Clio and the economics of Qwerty. The American Economic Review, 75: 332–337.Desai, M. A., Foley, C. F., Forbes, K. (2008) Financial constraints and growth: multinationaland local firm responses to currency depreciations. Review of Financial Studies, 21: 2857–2888.

Desai, M. A., Foley, C. F., Hines, J. R. (2004) A multinational perspective on capital structurechoice and internal capital markets. The Journal of Finance, 59: 2451–2487.

Egan, M. L., Mody, A. (1992) Buyer-seller links in export development. World Development, 20:321–334.

Fan, J. P. H., Lang, L. H. P. (2000) The measurement of relatedness: an application to corporatediversification. The Journal of Business, 73: 629–660.

Felipe, J., Kumar, U., Abdon, A., Bacate, M. (2012) Product complexity and economicdevelopment. Structural Change and Economic Dynamics, 23: 36–68.

Frenken, K., Van Oort, F., Verburg, T. (2007) Related variety, unrelated variety and regionaleconomic growth. Regional Studies, 41: 685–697.

Gaulier, G., Zignago, S. (2010) BACI: International Trade Database at the Product-Level, The1994–2007 Version. Working Paper, 2010–23. CEPII.

Gertler, M. S. (2003) Tacit knowledge and the economic geography of context, or the undefinabletacitness of being (there). Journal of Economic Geography, 3: 75–99.

Goh, A. T. (2005) Knowledge diffusion, input supplier’s technological effort and technologytransfer via vertical relationships. Journal of International Economics, 66: 527–540.

Goldberg, P., Khandelwal, A., Pavcnik, N., Topalova, P. (2009) Trade liberalization and newimported inputs. American Economic Review, 99: 494–500.

Gorodnichenko, Y., Schnitzer, M., (2010) Financial constraints and innovation: why poor countriesdon’t catch up, NBER Working Papers 15792. National Bureau of Economic Research.

Granovetter, M. (1973) The strength of weak ties. American Journal of Sociology, 78: 1360–1380.Grossman, G. M., Helpman, E. (1993) Innovation and Growth in the Global Economy, vol 1.Cambridge, MA: The MIT Press.

Hahn, C. H., Park, C. G. (2011) Direction of causality in innovation-exporting linkage: evidencefrom microdata on Korean manufacturing. Korea and the World Economy, 12: 367–398.

Hausmann, R., Hidalgo, C. (2011) The network structure of economic output. Journal ofEconomic Growth, 16: 309–342.

Hausmann, R., Hidalgo, C. A. (2009) The building blocks of economic complexity. Proceedingsof the National Academy of Sciences, 106: 10570–10575.

Hausmann, R., Hwang, J., Rodrik, D. (2007) What you export matters. Journal of EconomicGrowth, 12: 1–25.

Hausmann, R., Klinger, B. (2007) The structure of the product space and the evolution ofcomparative advantage, Working Paper 146. Center for International Development, HarvardUniversity.

Heimeriks, G., Boschma, R. (2014) The path- and place-dependent nature of scientific knowledgeproduction in biotech 1986–2008. Journal of Economic Geography, 14: 340–364.

Hidalgo, C. (2009) The dynamics of economic complexity and the product space over a 42 yearperiod, Working Papers 189. Center for International Development, Harvard University.

Hidalgo, C. A., Klinger, B. A., Barabasi, L., Hausmann, R. (2007) The product space conditionsthe development of nations. Science, 317: 482–487.

Howells, J. (2012) The geography of knowledge: never so close but never so apart. Journal ofEconomic Geography, 12: 1003–1020.

Jaffe, A. B., Trajtenberg, M., Henderson, R. (1993) Geographic localization of knowledgespillovers as evidenced by patent citations. The Quarterly Journal of Economics, 108: 577–598.

Krishnan, V., Ulrich, K. T. (2001) Product development decisions: a review of the literature.Management Science, 47: 1–21.

Lo Turco, A., Maggioni, D. (2015) Imports, exports and the firm product scope: evidence fromTurkey. The World Economy, 38: 984–1005.

Lundvall, B., Johnson, B. (1994) The learning economy. Journal of Industry Studies, 1: 23–42.Maskell, P., Malmberg, A. (1999) Localised learning and industrial competitiveness. CambridgeJournal of Economics, 23: 167–186.





ownloaded from


Melitz, M., Burstein, A. (forthcoming) Trade liberalization and firm dynamics. In R. Faini , J. deMelo, K. F. Zimmermann (eds) Advances in Economics and Econometrics: Theory andApplications, Econometric Society Monographs.

Neffke, F., Henning, M. S. (2008) Revealed relatedness: mapping industry space, Papers inEvolutionary Economic Geography (PEEG) 0819. Utrecht University, Section of EconomicGeography.

Neffke, F., Henning, M. (2013) Skill relatedness and firm diversification. Strategic ManagementJournal, 34: 297–316.

Neffke, F., Henning, M., Boschma, R. (2011) How do regions diversify over time? Industryrelatedness and the development of new growth paths in regions. Economic Geography, 87:237–265.

Neffke, F., Henning, M., Boschma, R. (2012) The impact of aging and technological relatednesson agglomeration externalities: a survival analysis. Journal of Economic Geography, 12: 485–517.

Nooteboom, B. (2000) Learning and Innovation in Organizations and Economies. Oxford: OxfordUniversity Press.

Orlando, M. (2000) On the importance of geographic and technological proximity for R&Dspillovers: an empirical investigation, Research Working Paper RWP 00-02. Federal ReserveBank of Kansas City.

Penrose, E. (1959) The Theory of the Growth of the Firm. New York: John Wiley.Perez, C., Soete, L. (1988) Catching up in technology: entry barriers and windows of opportunity.In G. Dosi, C. Freeman, R. R. Nelson, G. Silverberg, L. Soete (eds) Technical Change andEconomic Theory, pp. 458–479. London: Pinter.

Pfirrmann, O. (1994) The geography of innovation in small and medium-sized firms in westGermany. Small Business Economics, 6: 41–54.

Poncet, S., de Waldemar, F. S. (2012) Product relatedness and firm exports in China, WorkingPapers 2012–27. CEPII Research Center.

Porter, M. (2003) The economic performance of regions. Regional Studies, 37: 545–546.Romer, P. (1986a) Endogenous technological change. Journal of Political Economy, 98: S71–S102.

Romer, P. (1986b) Increasing returns and long-run growth. Journal of Political Economy, 94:1002–1037.

Salomon, R. M., Shaver, M. J. (2005) Learning by exporting: new insights from examining firminnovation. Journal of Economics & Management Strategy, 14: 431–460.

Sternberg, R., Arndt, O. (2001) The firm or the region: what determines the innovation behaviorof European firms. Economic Geography, 77: 364–382.

Teece, D. (2007) Explicating dynamic capabilities: the nature and microfoundations of(sustainable) enterprise performance. Strategic Management Journal, 28: 1319–1350.

Teece, D., Rumelt, R., Dosi, G., Winter, S. (1994) Understanding corporate coherence: theoryand evidence. Journal of Economic Behavior and Organization, 23: 1–30.

Timmermans, B., Boschma, R. (2014) The effect of intra- and inter-regional labour mobility onplant performance in Denmark: the significance of related labour flows. Journal of EconomicGeography, 14: 289–311.

Wang, C. C., Lin, G. (2012) Dynamics of innovation in a globalizing China: regionalenvironment, inter-firm relations and firm attributes. Journal of Economic Geography, 13: 397–418.

Wooldridge, J. (2002) Econometric Analysis of Cross-Section and Panel Data. Cambridge,Massachusetts: MIT Press.





ownloaded from


Appendix

Table A1. Variables: definition and description

Variable Description

dens Density indicator described in Equation (1)

densl Density indicator described in Equation (2)

Size Firm size measured as the log of the number of employees

LP Labour productivity measured as the log of real value added per worker

Exporter Exporter dummy equal to 1 if the firm exports in that year and 0 otherwise

Importer Importer dummy equal to 1 if the firm imports and 0 otherwise

Foreign Foreign ownership dummy equal to 1 if the firm is foreign owned

Multi-plant Multi-plant dummy equal to 1 if the firm has more than one production unit and 0

otherwise

R&D Share of R&D workers in total employment

w Log of average wage, calculated as the log of the total wage bill over number of

employees

Multi-product Multi-product dummy equal to 1 if the firm produces more than one product and 0

otherwise

RCAl RCA index. For each product is calculated as

yplPNj¼1 yjlPLl¼1 yplPL

l¼1

PNj¼1 yjl

wherein the formula L is the total number of NUTS 3 provinces in the Turkish economy,

and y denotes

the value of production

N Log of the number of products produced by the firm

NRCA l Log of the number of products with RCA in the province(s) where the firm is active

Logdens Log of density indicator described in Equation (1)

Logdensl Log of density indicator described in Equation (2)

yl Log of average local production value in the product calculated as:

Logyl ¼ Log½PLi

l¼1 slypl� withPLi

l¼1 sl ¼ 1 8i

denslRCAWorld

Local density measure calculated as in Equation (2), but with the set of RCA products

selected on the basis of RCA ¼

yplPNj¼1 yjl

WorldExportspTotalWorldExports

denslRCA05

Average local density measure calculated as in Equation (2), but with

the set of RCA products selected on the basis of RCA40.5

densNorm Firm density measure from Equation (1) divided by the total number of

products that the firm produces

denslNorm Average local density measure where local densities in Equation (2) are

divided by the number of RCA products in each NUTS 3 Province

densR Average local density measure calculated in Equation (2)

on the basis of NUTS 2 regions locations

RCAR RCA index. For each product is calculated as

yprPNj¼1 yjrPRr¼1 yprPR

r¼1

PNj¼1 yjr

wherein the formula R is the total number of NUTS 2 regions in the Turkish economy,

and y denotes the value of production





ownloaded from


Table

A2.

Robustness:

alternativediversificationdefinitions

[1]

[2]

[3]

[4]

[5]

[6]

[7]

[8]

Conservativediversification

Radicaldiversification

within

producedtw

o-digit

codes

outsideproducedtw

o-digit

codes

Random

selectionof

Exclusionof

Exclusionofproduced

2006

2007

2008

2009

Random

selectionof

10%

ofzeroes

new

two-digit

codes

four-digit

codes

10%

ofzeroes

densl

0.247***

0.030***

0.014***

0.009***

0.007***

0.011***

0.011***

0.059***

[0.049]

[0.006]

[0.005]

[0.001]

[0.001]

[0.002]

[0.002]

[0.010]

dens

6.092***

1.953***

0.787***

2.610***

2.605***

2.496***

2.194***

13.677***

[0.782]

[0.184]

[0.158]

[0.079]

[0.081]

[0.106]

[0.085]

[0.257]

Size

0.007

0.001

0�0.003*

[0.008]

[0.001]

[0.001]

[0.001]

LP

�0.003

00

0

[0.004]

[0.001]

[0.000]

[0.001]

Importer

�0.009

�0.001

�0.001

�0.003**

[0.008]

[0.001]

[0.001]

[0.001]

Exporter

0.004

00.001*

0.001

[0.008]

[0.001]

[0.001]

[0.001]

Foreign

0.023

0.009**

0.006

0.022**

[0.031]

[0.004]

[0.006]

[0.009]

Multi-plant

0.003

0.002

00.001

[0.008]

[0.001]

[0.001]

[0.002]

RCAl

0.006***

0.001***

0.001***

0.000***

0.000***

0.000***

0.000***

0.002***

[0.000]

[0.000]

[0.000]

[0.000]

[0.000]

[0.000]

[0.000]

[0.000]

FE:

Year

yy

yy

yy

yy

Product

yy

yy

yy

yy

Firm

yy

yy

yy

yy

Observations

104,256

893,401

762,795

4,531,279

3,109,521

2,499,714

2,795,065

1,190,127

R2

0.338

0.07

0.058

0.038

0.04

0.042

0.039

0.144

�p<

0:10;��p<

0:05;��p<

0:01.Dependentvariable:firm

probabilityto

introduce

new

product

p.

Standard

errors

are

clustered

byfirm

.Allregressors

included

intheestimationare

1-yearlagsofthecorrespondingvariables.





ownloaded from


Table A3. Robustness: alternative modelling choice

[1] [2] [3] [4] [5] [6] [7]

Conditional logit Logit RE Logit Probit

2006 2007 2008 2009

dens 24.904*** 26.243*** 18.504*** 13.972*** 75.445*** 75.447*** 34.051***

[2.449] [2.887] [3.444] [3.315] [1.722] [1.722] [0.756]

densl �0.161 0.279** 0.538*** 0.228* 0.808*** 0.808*** 0.310***

[0.108] [0.131] [0.141] [0.129] [0.080] [0.080] [0.031]

Size 0.007 0.031 0.028 0.002 0.039*** 0.039*** 0.014**

[0.016] [0.022] [0.025] [0.022] [0.015] [0.015] [0.006]

LP 0.006 �0.015 �0.043* �0.047* �0.075*** �0.075*** �0.030***

[0.016] [0.023] [0.025] [0.025] [0.014] [0.014] [0.006]

Importer �0.015 0.061 0.022 0.009 �0.108*** �0.108*** �0.043***

[0.036] [0.041] [0.045] [0.044] [0.028] [0.028] [0.011]

Exporter �0.022 �0.022 0.041 0.003 �0.056** �0.056** �0.024**

[0.032] [0.037] [0.043] [0.041] [0.025] [0.025] [0.010]

Foreign �0.086 0.242** �0.026 0.021 �0.051 �0.05 �0.013

[0.085] [0.097] [0.134] [0.130] [0.064] [0.064] [0.026]

Multi-plant �0.036 �0.043 �0.018 �0.042 0.03 0.03 0.013

[0.031] [0.036] [0.040] [0.039] [0.024] [0.024] [0.010]

RCAl 0.039*** 0.035*** 0.040*** 0.035*** 0.034*** 0.034*** 0.018***

[0.002] [0.002] [0.002] [0.002] [0.001] [0.001] [0.000]

Observations 225,853 135,002 99,357 122,709 897,344 897,344 897,344

�p < 0:10;��p < 0:05;��p < 0:01. Standard errors are clustered by firm. All regressors included in the

estimation are lagged to 1 year. Year fixed effects are included in all specifications.





ownloaded from


A - Production Value in 2005

B - Average Production Value Growth 2005/2009

C -Average weight of New Products in Production Value - 2005/2009

Figure A1. Turkish manufacturing production (2005–2009). (A) Production value in 2005; (B)average production value growth (2005–2009); (C) average weight of new products inproduction value (2005–2009).Notes: Quartiles of variables distribution are represented by means of different grey tonalities,with the darker ones identifying upper quartiles.The top panel displays the NUTS 3 spatial distribution of Turkish manufacturing productionvalue. The middle-panel chart displays the NUTS 3 spatial distribution of Turkishmanufacturing production value average growth. The lower-panel chart displays the NUTS 3spatial distribution of the 2005–2009 average weight of new products in manufacturingproduction value.Source: TurkStat SBS and AIPS (own calculations).





ownloaded from


Figure A2. Relative contribution of densNotes: Quartiles of variables distribution are represented by means of different grey tonalities,with the darker ones identifying upper quartiles.The map displays the NUTS 3 spatial distribution of the relative importance of firm density in

explaining NUTS 3 region average innovation rates, Iipt:a2�dens

a1�denslþa2�dens.

Source: TurkStat SBS and AIPS (own calculations).





ownloaded from


on firms’ product space evolution: the role of firm and

Documents